DragAnything: Motion Control for Anything using Entity Representation
This work addresses the challenge of user-friendly motion control in video generation for applications requiring interactive object manipulation, though it is incremental in building on existing trajectory-based methods.
The paper tackles the problem of controllable video generation by introducing DragAnything, a method that uses entity representation to enable motion control for any object, achieving state-of-the-art performance with a 26% improvement in human voting over previous methods.
We introduce DragAnything, which utilizes a entity representation to achieve motion control for any object in controllable video generation. Comparison to existing motion control methods, DragAnything offers several advantages. Firstly, trajectory-based is more userfriendly for interaction, when acquiring other guidance signals (e.g., masks, depth maps) is labor-intensive. Users only need to draw a line (trajectory) during interaction. Secondly, our entity representation serves as an open-domain embedding capable of representing any object, enabling the control of motion for diverse entities, including background. Lastly, our entity representation allows simultaneous and distinct motion control for multiple objects. Extensive experiments demonstrate that our DragAnything achieves state-of-the-art performance for FVD, FID, and User Study, particularly in terms of object motion control, where our method surpasses the previous methods (e.g., DragNUWA) by 26% in human voting.